How good is it at correcting things you point out directly? I haven’t used it for coding yet but have noticed it’s ok at correcting mistakes when you point them out. Still hit or miss though.
Comment on Solve a puzzle for me
psud@aussie.zone 7 months agoYeah. I asked GPT3 for some heliostat code, to keep reflected sunlight stationary. It was wrong, it hallucinated libraries that didn’t exist, but it roughed out a program that was easier to fix than it would have been to start from scratch.
Maybe its superpower is beating inertia, getting you started
Buddahriffic@lemmy.world 7 months ago
psud@aussie.zone 7 months ago
It was ok. I and it went through about four iterations going from “that’s a sun tracker, I asked for a heliostat” through undeclared variables, global variables that should have been local until it was a fine program with just the fault that there was no such library as solar::heliostat [azimuth, altitude]
I have read that people have run into that sort of problem and have written the library the AI called for, but I looked up a real astronomy library
Dultas@lemmy.world 7 months ago
Yeah, I’ve used it for a lot for one off data processing / graphing code, stuff that is to big to process in a spreadsheet. It usually gets like 95% there. The real issue I have is if you ask for too many one off adjustments it gets confused and reverts previous changes when you ask it to make new ones.